WAYNE STATE UNIVERSITY (BIG) DATA DRIVEN ANALYTICS: Enhancing Emergency Healthcare Ratna Babu Chinnam, Ph.D. STRENGTHENING THE BIG DATA & ANALYTICS ECOSYSTEM WAYNE STATE UNIVERSITY September 19, 2018 Co-Director, Big Data & Business Analytics Group Professor, Industrial & Systems Engineering Lead, Business Analytics Spoke, NSF s Midwest Big Data Innovation Hub Ratna.Chinnam@wayne.edu 313.577.4846 bigdata.wayne.edu COLLABORATORS: Dr. Seth Krupp & Dr. Mike Nauss, ED Directors, HFHS Dr. Seung Yup Lee & Dr. Evrim Dalkiran, ISE Department
Why Big Data Analytics/Technologies? DOMAINS OF INTEREST TO PRIVATE SECTOR MY EXPERIENCE Customer transactions 73% Customer/prospect data 56 Market & competitive data Service data Warranty & fraud detection Channel data 49 49 49 44 Product data 34 Industry specific data 24 Supply chain data 12 SOURCE: NEWVANTAGE PARTNERS, BIG DATA EXECUTIVE SURVEY 2012-2018 & HBR.ORG STRENGTHENING THE BIG DATA & ANALYTICS ECOSYSTEM R.B. CHINNAM 19 SEPTEMBER 2018 2
ED Visits (Millions) Case Study: Enhancing Emergency Healthcare Current State: Ineffective service and a national crisis! 2015: 136.9 Million Visits (CDC) AHA Survey of Community Hospitals in U.S. (2015) Opportunity: Growing adoption of EHR systems in Hospitals Year: 2008 Number of EDs Medical Errors: 250k Deaths/Yr 76% due to Information Processing / Verification (Schnapp et al. 2018 LINK) 500 400 300 200 100 0 AK AL AR AZ CA CO CT DC DE FL GE HA IO ID IL IN KA KE LO MA MD ME MI MN MO MS MT NC ND NE NH NJ NM NV NE OH OK OR PE PR RI SC SD TE TX UT VI VE WA WI WV WY Median Boarding Time (Admitted s) in Minutes 2011 Data Source: CMS / Propublica.org Data Span: 2016 Median Time for : 5.5 hours Median Boarding Time: 2.25 hours Median Time Until Sent Home in Minutes (Discharged s) 2015 Adoption: 9% 28% 84% Source: Office of the National Coordination for Health Information Technology Approach: Data and AI to improve real-time operational intelligence for enhanced proactive orchestration of healthcare operations! STRENGTHENING THE BIG DATA & ANALYTICS ECOSYSTEM R.B. CHINNAM 19 SEPTEMBER 2018 3
Typical ED Care Giving Process Resuscitation Acute Care Unit Fast Track Arrival Triage Waiting Room Room Pediatric Psychiatric Other Treatment Physician Assessment Follow Up Treatment BOARDING DELAY FOR ADMITTED PATIENTS 24 mins 12 mins 23 mins 220 mins 258 mins Time Stats from Henry Ford Hospital (May 2014 -Dec 2016) Departure from ED Approval Preparation Transfer Main driver of overcrowding 48% of ED length of stay CMS requiring boarding stats (2014) STRENGTHENING THE BIG DATA & ANALYTICS ECOSYSTEM R.B. CHINNAM 19 SEPTEMBER 2018 4
Number of patients General Wisdom ED IU Allocation Delay (BAD) 28% 15% 15% 32% 38% 21% 0AM 0 2 4AM 4 68AM 8 10 12PM 14 4PM 16 18 8PM 20 0AM 22 Hour of day BAD: --- >2Hours --- >6Hours --- >10Hours Number of patients 18 16 14 12 10 Pattern 8 suggests that 6 boarding delays 4 are probably not 2due to full IU occupancy 0 ED is Congested with Boarded s for Lack of Inpatient Unit (IU) s! IU Rate Patterns 0AM 0 2 4AM 4 6 8AM 8 1012PM 144PM 16 18 8PM 20 0AM 22 ED s Discharge Hour of day All admissions Probability Full IU Occupancy Probability (No beds Zero or one unoccupied beds) 0.7 Boarding delays attributable to improper 0.6 0.5 coordination within the 0.4 0.3 Period of Peak s from ED ED-IU network! 0.2 Adding staff not the 0.1 solution. Need to know 0 which 0AM 0 2 4AM 4 6 8AM IU 8 1012PM beds 14PM 16 18 to 8PM 20 0AM 22 Hour of day turnaround. Internal Med 1 Internal Med 2 Pulmonary Nephrology HEM/ONC/BMT Orthopedics INSIGHT: Predict IU admissions from ED and timing to facilitate proactive coordination of downstream resources/processes! STRENGTHENING THE BIG DATA & ANALYTICS ECOSYSTEM R.B. CHINNAM 19 SEPTEMBER 2018 5
Arrival Triage Proactive IU Reservations : Modelling Physician Examination FORK-JOIN QUEUEING MODEL REPRESENTATION λλ ee 1 pp Emergency Department No NOTATION: Resource Departure from ED to unit other than inpatient unit ωω Yes pp Seung Yup Lee, Ph.D., Ratna Babu Chinnam, Ph.D., Evrim Dalkiran, Ph.D. 1 rr 1 qq rr Fork ss 1 qq Remaining processes in ED (M/M(μμ 11 )/ ) λλ: patient flow rates qq: true positive probability (classifier) μμ 1 : lead-time for proactive bed request signal (decision variable) Fork from other than ED λλ oo Fork ss 2 preparation (M/M(μμ 22 )/s) pp: probability of sending bed request rr: false negative probability (classifier) μμ 2 : service time at bed preparation server Join (bed assignment) ss 3 Arrival to unit ωω through physical transfer of patients Minutes EXPERIMENT SETTING & RESULT: 100 90 80 70 60 50 40 General IU & Imperfect Disposition Predictions Inclusion of s Being Admitted from Other Sources Rate to a Single IU: 0.2 s/hour Each from the ED and Other Sources (50:50) Assumption: Unbiased disposition prediction and remaining ED LoS BAD with reactive bed preparation of 1hr >30% reduction in BAD ED patients >50% reduction in BAD! Non-ED patients 0 0.25 0.5 0.75 1 Precision (Predictions) INFORMS SERVICE SCIENCE BEST PAPER AWARD 2017 STRENGTHENING THE BIG DATA & ANALYTICS ECOSYSTEM R.B. CHINNAM 19 SEPTEMBER 2018 6
AI Powered Predictive & Prescriptive Analytics Arrival Triage Encounter with Doctor Lab / Imaging Results ED Process Flow PROVIDER NOTES Electronic Health Record ANALYTICS DEVELOPMENT: Prediction Models: Deep Learning using TensorFlow & NLP Explainable AI: Gradient & Perturbation Attribution Methods Prescriptive Analytics: Proactive Coordination Signals Growing information with care (structured & unstructured data) to power predictions! RESULTS: 225k s >6M Text Notes >5M Lab/Imaging Results >90% Disposition Accuracy STRENGTHENING THE BIG DATA & ANALYTICS ECOSYSTEM R.B. CHINNAM 19 SEPTEMBER 2018 7
Impact on ED Processes at Henry Ford Hospital ORIGINAL STATE: Approval Preparation Transfer Significant Boarding PHASE #1: Parallel bed preparation during admission approval Approval Preparation Transfer Reduced Boarding PHASE #2: Parallel bed preparation during ED treatment ED Processes Preparation Approval Transfer Minimal Boarding Well executed Big Data Analytics can have remarkable impacts! STRENGTHENING THE BIG DATA & ANALYTICS ECOSYSTEM R.B. CHINNAM 19 SEPTEMBER 2018 8